- Amazon internally codenamed its cancellation flow "Iliad"—after Homer's epic about a long, brutal war.
They designed a 4-page, 6-click, 15-option maze to stop you from leaving.
The FTC noticed. Every enterprise should pay attention.
🧵 - The "Click-to-Cancel" rule was finalized Oct 2024. Simple mandate: cancelling must be as easy as signing up.
The Eighth Circuit vacated it in July 2025 on procedural grounds.
But the underlying enforcement pressure hasn't gone anywhere. - Amazon's "Iliad Flow" is a masterclass in dark patterns:
— "Keep my benefits" in bright blue
— "Continue to cancel" in muted gray
— Marketing pages, offer pages, then finally cancellation
Designed to exhaust you into giving up. - Epic Games paid $245M—the largest FTC administrative settlement ever.
One-button purchases with no confirmation. Kids racking up hundreds on parents' cards.
Then Epic locked accounts of users who disputed charges. - Now it's getting worse. Enterprises are deploying AI "save agents"—LLM wrappers with one objective: prevent cancellation at all costs.
These agents use emotional manipulation, guilt anchoring, and relentless nagging disguised as conversation. - Research shows AI dark patterns are more embedded and subtle than visual ones.
An agent might reference your personal data mid-cancellation: "How's recovery from your surgery going?"
Rapport weaponized as a retention tool. - At Veriprajna, we think this entire approach is broken.
Predicting churn ≠ preventing churn. Most "save" flows waste resources on users who'll leave anyway—or worse, annoy loyal users into leaving.
Our approach uses Causal AI and Uplift Modeling to answer a different question: - Not "who will churn?" but "who will ONLY stay if we intervene?"
That distinction changes everything.
It segments users into four groups: - — Persuadables: intervene here (only real ROI)
— Sure Things: skip the discount, they're staying
— Lost Causes: let them go gracefully
— Sleeping Dogs: contact them and they'll cancel - We align retention agents with multi-objective RLHF—not optimized for "prevent cancellation" but for clarity, helpfulness, and zero shaming.
If the agent can't persuade within defined steps, it surfaces one-click cancel immediately. - The enterprises that win long-term won't be the ones with the most elaborate cancellation mazes.
They'll be the ones customers trust enough to return to.
Should AI retention agents be required to disclose they're optimizing for retention—not helping you? #AIethics #RetentionAI - We wrote the full deep dive—covering the FTC landscape, causal inference math, and RLHF alignment frameworks:
https://veriprajna.com/whitepapers/ethical-frontier-retention-engineering-algorithmic-accountability